Optimal Scheduling Strategy For Multi Energy Security Based On Multi Agent Reinforcement Learning Algorithm

With the deepening of human industrialization,the dependence on fossil fuels has been increasing.Scientists have discovered that due to human overuse of fossil fuels,the environment essential for survival has become increasingly severe.Therefore,people have begun to pay attention to the use of intermittent renewable energy sources.Although intermittent energy sources such as wind and solar energy are abundant,they are influenced by uncontrollable factors such as weather and seasons.As a result,intermittent energy sources exhibit volatility and uncertainty,posing significant challenges to the secure operation of power grids.In response,numerous experts at home and abroad have conducted in-depth research,and this article provides a detailed analysis from the perspectives of load forecasting and demand response.Firstly,considering the uncertainties caused by different user habits in terms of load,it is proposed to use load forecasting methods to ensure power economic benefits in advance and maximize the balance between power supply and demand.This article introduces a Convolutional Neural Network-Long Short-Term Memory(CNN-LSTM)approach for load forecasting.The algorithm analyzes specific load features and applies data cleaning,smoothing,and standardization techniques for data preprocessing.Based on this,the CNN-LSTM model,which combines Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM),is employed for load forecasting.The CNN is responsible for feature extraction,and compared to LSTM,it effectively reduces the dimensionality of input data,thereby reducing the difficulty of prediction and improving load forecasting accuracy.Simulation comparisons show that the CNN-LSTM algorithm outperforms the BP and LSTM algorithms in terms of prediction accuracy and training speed.Secondly,considering the large and irregular demand of the demand-side load,it is proposed to use dynamic pricing as an incentive to adjust the elastic load of electricity users,optimize their electricity usage strategies,and ensure the economic benefits of the power grid.A dual-side power dispatching demand response model,constructed by both the power supplier and the user,is established.To optimize this model,the Multi-Agent Deep Deterministic Policy Gradient(MADDPG)algorithm is employed.Unlike traditional algorithms,MADDPG can handle continuous action spaces,enabling it to tackle a wider range of problems.Moreover,the algorithm utilizes a centralized training Critic network and decentralized training Actor networks,which allows for collaboration and enhances the overall performance of the agents.The case study demonstrates the effectiveness of MADDPG in the demand response model,achieving the effect of load peak shaving and valley filling.In conclusion,this article proposes the utilization of CNN-LSTM for load forecasting and the adoption of the MADDPG algorithm for demand response,addressing the challenges posed by the volatility and uncertainty of intermittent renewable energy sources.These approaches demonstrate superior performance in terms of load forecasting accuracy,training speed,and the achievement of load peak shaving and valley filling in demand response.